On November 10, 2026, Google Cloud introduced Gemini for healthcare — an industry-specific agentic AI solution that unifies disparate clinical, operational, and administrative medical systems into a single managed layer and executes multi-step workflows: medical record synthesis, patient visit preparation, insurance coverage checks, and prior authorization. The product is in preview mode and is currently available only in the US, with the agent connecting to electronic health records exclusively in read-only mode.

image
image
image

What happened

The platform connects electronic health records (EHR), billing, clinical guidelines, and Medicare data into a single managed layer and processes multi-step workflows on it: medical record synthesis, patient visit preparation, coverage checks, and prior authorization, which is the pre-approval of treatment with the insurance provider. EHR integration works without write permissions: Epic via open.epic, Oracle Health via Ignite/FHIR. In addition, the announcement lists connectors to InterSystems FHIR, Verato, and Keragon, which claims support for more than 400 applications, plus access to CMS data. Compliance measures are listed separately: HIPAA support, role-based access control (RBAC), and verifiable citations as a mechanism for answer traceability. At the same time, an "open model choice" — Google, Anthropic, and OpenAI — is announced, with smart routing of tasks between fast and highly reasoning models. With launch partners HCA Healthcare, Highmark Health, and Northwell Health, the solution is deployed in the US in preview mode.

Context

The importance here lies not in a new model, but in the scale of the problem the platform addresses. According to a Harris Poll survey cited by Google, doctors spend about 28 hours a week on administrative paperwork, and according to estimates by the American Hospital Association (AHA), administrative costs account for 25 to 35 percent of US healthcare spending. Medical data is fragmented: medical records are stored in EHR, payments live in billing, and coverage rules are in insurer documentation, so a unified agentic loop connecting these systems has not existed until now. Gemini for healthcare is Google's first major "agentic" product for medicine, and the company is betting not on a separate model, but on an integration layer, ready-made skills, and access control.

Why this matters for the industry

For the industry, the main signal is the architecture: this is a reference template for agentic AI in a regulated industry, where the connector layer to Epic, Oracle Health, InterSystems FHIR, Verato, and Keragon is combined with ready-made administrative process skills, an open model choice under a unified governance layer, and answer citability. The "read-only" mode legitimizes read-only agents as a reasonable first step of autonomy in systems where the cost of error is high, and support for models from three vendors sets the standard for a model-agnostic enterprise agent. Practical consequences: startups and vendors of agentic AI in medicine are re-evaluating their positioning against Google's ready-made connector layer, hospitals gain negotiating leverage when choosing suppliers, and builders can use the announcement as a checklist when designing their own agentic systems: skills, connectors, open model choice, citability, and RBAC.

Why this matters for users

For readers in healthtech, this is a benchmark for architecture and ecosystem: Suki, UiPath, Alteryx, GoodRx, and Infinitus Relay have already gathered around the platform, and their integrations show which scenarios the market considers the first candidates for automation. There is no direct access in preview: the product is geo-restricted to the US and distributed through launch partners, so an external team cannot test it right now. The effect for patients is delayed: if the platform proves its viability, prior authorization and visit preparation become faster, and doctors free up time for patients instead of paperwork, however, no confirmed measurements from pilot results have been published.

What is still unknown / limitations

There are no quantitative confirmations of quality: the announcement lacks benchmarks, error rates, and citation accuracy metrics for prior authorization and medical record synthesis, and all three publications are Google's own materials without independent verification; launch partners have also not published pilot results. The criteria for smart routing between vendor models, behavior during failures, and error attribution are not disclosed, and the cost has not been announced. Verifiable citations solve the attribution question but do not guarantee the correctness of the decision: an accurate citation may be incorrectly applied to an insurance coverage rule, so citation precision and recall metrics and a public breakdown of failure modes are needed. Finally, the question of write access to EHR and whether the platform will expand beyond the US remains open.

Sources

Author

Look at AI, editorial team